Low noise differential 4:1 Demux with EMI filtering?

On a sunny day (Tue, 31 Mar 2015 10:21:50 +0200) it happened Rune wrote in :

What is you input frequency range? If low you could just decouple the inputs with some C or LC to ground.

Input frequency range is 0 to 700 kHz, filtered on the amplifier inputs after the demux. The RF immunity requirements ranges from 80 to 2.7GHz. A LC filter or typical EMI filter package is a valid and common solution, but rater space consuming in this application since there are

64 input lines. I am hoping not to increase the component count on this board any further.

Thanks anyway.

How about one RFI filter after the MUX? Its cutoff freq could be high enough not to slow the channel-switching response.

--
 Thanks, 
    - Win

On a sunny day (31 Mar 2015 03:08:25 -0700) it happened Winfield Hill wrote in :

I agree with that. There was a thread here some time ago about RF detection by some opamps. So changing opamp maybe a possibility too.

I have not had RF detection in analog muxes of the CMOS kind.

So it is worth checking where the RF detection takes place.

In our old Lab in downtown San Jose, CA I used to routinely watch on a scope over 15mVpp of AM Radio to 'noise' appear on lines. Interestingly you could easily tell when the DJ was talking or the station was playing music. But suffice it to say MAJOR NUISANCE!!

Today you have to watchout for Security Guards' walkie talkies that blow over 1W of power out as they walk by AND there're GSM telephones, AND there're bluetooth band etc. ALL can put little spikes and bursts into your front end. You MUST determine what RF source you're sensitive to, You may have an RF 'bandpass' sensitivity that exacerbates your problem. Again, find what RF source is causing the problem. As a simple test, bring an RF Generator near by using a small wire antenna, modulate the RF with

1kHz AM/FM tone and look for the spike in your ADC's output. SThe sensitivity is probably in the 1 to 1.5 MHz RF radio station range is my guess.

That said, How to fix WITHOUT adding components at64 lines. Go back through your PCB layout and make certain you have BALANCED input lines. UNBLANCED inputs change the RF into differential and that makes the voltages on the line look like 'real' signals. To improve, you may still need to add Common Mode chokes on each pair coming in from the outside world, though. At least they're broadband, small SMD's.

If the MUX detects RF, no hope for post filtering. May work enough if the MUX is NOT detecting and passing the energy through. A sharp cutoff LPF may give you enough. When you work in this stuff, it never goes away, you just lower it. You didn't say how much this EMI is raising your noise floor.

You mentioned 64 differential inputs. Hmmm. Then you can't balance the input lines, You're stuck with removing the energy early. At the input lines themselves, or maybe get lucky, after the MUX. Have you looked at SoundCard layouts to see what they do? Take a look at Creative Labs EMU1212, EMU0404, input areas. It's few channels, but they get 22 bits of 'clean' signals.

That would work if the demodulation/intermod/whatever isn't happening in the MUX.

It's probably worth hacking up the board a bit to find out -- if it were me I'd try putting some RF filtering in front of one MUX, and some RF filtering behind another, and then see if there's a difference in the amount of problems on the relevant channels.

If you're filtering behind a MUX you have to make sure the filter is opened up enough to accommodate sampling; sometimes that by itself rules out filtering. It depends on the problem frequencies and the sampling rate, o-course.

--

Tim Wescott 
Wescott Design Services 
http://www.wescottdesign.com

Ferrite beads have been mentioned; depending on what type of connectors and pinouts you're using, maybe one of those multi-hole plates would be useful. Like,

I think they're usually poorly stocked though.

On the upside, since you say your signals are mostly low impedance, then your circuit should also be low impedance, making filtering easy. Unless it's not, especially at higher frequencies, in which case you'll need capacitors on the board, too; simply no way around that. You can get FB and MLCC arrays to facilitate that, at least.

As for mechanism, it's not apparent what amplitude and frequency is being tested with. Most tests are done in the 10V/m range, so if your cable is about 1m long, figure on about 10V appearing... most anywhere in the circuit. More than enough to blow out logic thresholds and low voltage analog signals, so you can get RF slapping into positive and negative input protection diodes, and stuff like that.

And down in the microvolts, yeah, it could very well be nonlinearity of the switch itself -- Rds(on) varies with Vcm (should be in the datasheet? -- haven't checked it myself), so you get a second order modulation effect due to common mode voltage. The switch will also attenuate some (due to distributed Rds(on) vs. Cdss), but might be nonlinear in the process, for the same reasons.

And as Bob said, asymmetry in the circuit can do stuff. And I suppose out of microvolts, it still needn't take much.

Are you using bipolar op-amps, by the way?

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com 

"Rune"  wrote in message  
news:mfdler$je2$1@speranza.aioe.org... 
> Hi all, 
> 
> We have a data acquisition board with an 8 channel ADC and low noise  
> amplifier front ends. In front of these amplifiers I have placed 8 4:1  
> differential analog demultiplexers, ADG1409 by the way, so that I am  
> able to measure a total of 32 differential analog inputs. Anyway we are  
> experiencing some RF noise demodulation problems with this setup,  
> probably in the demultiplexers. The most efficient way I can think of to  
> reduce the problem, in regard to cost and space budget, is to replace  
> the demuxes with some that have better EMI performance, probably with  
> some sort of built in EMI filters, if such a thing even exist? 
> 
> So any suggestions are much appreciated. The demultiplexer does operate  
> from a +-7V supply. And low noise is important therefore the on  
> resistance should preferably be < 20 ohm. 
> 
> Thanks, and kind regards. 
> 
> Rune